Developing A New Adaptive Optimal k-Nearest Neighbor Methodology for Flight Test Data Anomaly Detection – Application to Business Aircraft

The advancement of flight data analysis algorithms for improving the operational safety and efficiency of the aviation industry is always vital. This paper presents a novel adaptive optimal k-Nearest Neighbor (kNN) algorithm designed to detect anomalies in flight test data. This enhanced methodology addresses the limitations of traditional kNN algorithms by optimizing number of neighbours and designing an adaptive threshold mechanism that dynamically adjusts to the noise and outlier characteristics inherent in-flight data. The proposed approach not only improves the detection accuracy but also adapts to the changing dynamics of flight data, ensuring high sensitivity and specificity in anomaly identification. Through rigorous testing on longitudinal trim condition data, the algorithm demonstrates very good performance in recognizing spikes and failures that could indicate potential safety risks.